HackStack

Project Info

Co:Lab

Devpost

Inspiration

Agentic intelligence is rewriting the rules of knowledge work. By dissolving intelligence bottlenecks in fields like software and finance, AI agents are decoupling general productivity from human availability. Inspired by this propagation of intelligence across digital industries, we identified the physical sciences (e.g. Biology, Chemistry) as a next frontier for agentic integration. Researchers today are stalled by rote, manual tasks: running physical experiments, parsing fragmented data, or pipetting. In many cases, these low-leverage procedural tasks can consume over 40% of a researcher’s time, slowing discovery. We envision a solution which bridges the gap between LLMs and lab hardware, closing the loop between experiment design, execution, and iteration and unlocking scientists to focus on high-leverage research direction and novel work.

What it does

Co:Lab is an AI integrated laboratory that accelerates experimental design, execution, and iteration. Our platform enables researchers to focus more on high-level research direction and novel investigations, rather than rote labor. Our system has three core features: Researcher-in-the-loop: We believe that effective agentic systems know when to request human confirmation, especially in high-impact, technical environments. Rather than being a black box, Co:Lab’s interface enables researchers to confirm experiment design and observe experiment progress, with the ability to jump in if needed. Real-time observability: Experiments should be highly visible, both to the agent and the researcher. Our integrated system has a variety of cameras, sensors, and measurement devices which collect data that we pass to the user interface. This enables researchers and agents to get on-demand visibility into experiment state. Further, we believe in a future where researchers will commonly run concurrent experiments, making effective observability even more critical. Ability to iterate: Being able to perform one experiment isn’t enough. Scientific discovery and confidence is built by iterative experiments — refining hypotheses, adjusting parameters, validating results — which uncover new information or confirm prior assumptions. Co:Lab’s agentic loop enables it to analyze prior experiments and incorporate external research to propose subsequent experiments. This enables it to make well-informed, high-confidence decisions about what to test next. Altogether, Co:Lab bridges existing LLM capability with non-agentic lab benches to create an adaptive, agentic laboratory — one that thinks, runs, and improves experiments alongside researchers.

How we built it

Our project implements a comprehensive system architecture that combines our hardware, user interface, and agentic harness. We created a system diagram to demonstrate the information flow between each component. Hardware: Our autonomous laboratory is built using standard laboratory instruments: Servos Raspberry Pi 3D printed components Pulse width modulator Frontend: Our frontend uses React, Next.js, Typescript, Tailwind and Shadcn for a performant, modern application. Backend: Our backend API layer is written in Python, facilitating realtime communication between our frontend, hardware, and agents using websockets.

Challenges we ran into

Transforming standard laboratory instruments into an autonomous lab Integrating hardware, frontend, and backend together as a two-person team Fine tuning agentic loops

Accomplishments we're proud of

Working with microcontrollers and electronics to build a custom autonomous lab Demonstrating how AI can accelerate iteration loops

What we learned

Real-time integration of hardware into frontend interfaces How to design for an experiment, both in terms of experiment design and in terms of architecting the laboratory itself

What's next

Expand and modularize lab setup to support more systems and use cases Further optimize and fine-tune agent harnesses (e.g. better web search) for lab work

Analysis

Compare with all teams

View

Metric

Figures cover GitHub contributors during the hackathon window. A co-authored commit counts in full for each author, so per-member totals add up to more than the whole-team figures.

Technology

Found in codeClaimed only
  • AnthropicIn code
  • CSSIn code
  • JavaScriptIn code
  • Next.jsIn code
  • PythonIn code
  • ReactIn code
  • Tailwind CSSIn code
  • TypeScriptIn code
  • Vercel AI SDKIn code
  • VercelClaimed

9 of 10 appear in the indexed code. 1 claimed on Devpost could not be matched to code, which may simply mean the tool leaves no trace in the repository.

AI coding agents

  • CursorConfig · Commits

Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.

Codebase size

Source size

1.4 MB

Source files

142

Counts recognized source files only; vendored directories, binaries and lockfiles are excluded, so this is smaller than the repository on disk.

0 stars